Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Transfer Reinforcement Learning Using Output-Gated Working Memory
Arthur Williams, Joshua Phillips
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| pretrain on Red-Green source task | output gate model · Red-Green-Purple gridworld time to threshold | time to threshold: 32k -> 16.18k | Improved |
| pretrain on Red-Purple source task | output gate model · Red-Green-Purple gridworld time to threshold | time to threshold: 32k -> 19.76k | Improved |
| pretrain on Red source task | output gate model · Red-Green-Purple gridworld time to threshold | time to threshold: 32k -> 35.74k | Hurt |
| pretrain on Purple source task | output gate model · Red-Green-Purple gridworld time to threshold | time to threshold: 32k -> 1 | Improved |
| pretrain on Red-Green source task | input gate model · Red-Green-Purple gridworld time to threshold | time to threshold: 3.46k -> 0.83k | Improved |
| pretrain on Red-Purple source task | input gate model · Red-Green-Purple gridworld time to threshold | time to threshold: 3.46k -> 1.11k | Improved |
| pretrain on Red source task | input gate model · Red-Green-Purple gridworld time to threshold | time to threshold: 3.46k -> 2.37k | Improved |
| pretrain on Purple source task | input gate model · Red-Green-Purple gridworld time to threshold | time to threshold: 3.46k -> 2.75k | Improved |
| use source-task transfer (jumpstart) | output gate and input gate models · Red-Green-Purple gridworld time to threshold | initial error (jumpstart) | Improved |
| use output gating instead of input gating for transfer | working memory agent · Red-Green-Purple gridworld time to threshold | jumpstart differential | Improved |
| allow input and output gates full autonomy to open or close regardless of environment (ablation) | gated working memory model · grid-world maze | task solving | Hurt |
| constrain input gate to open only when environment presents features available for storage (ablation) | gated working memory model · 5x5 grid-world maze (25 states) | task solving | Improved |
Closest papers by meaning
- Model-Based Transfer Learning for Contextual Reinforcement Learning (NeurIPS 2024 · similarity 0.85)
- SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training (ICML 2025 · similarity 0.85)
- On the Importance of Gating: Memorization vs. In-Context Learning in State Space Models (NeurIPS 2026 · similarity 0.84)
- Think Before You Act: Decision Transformers with Working Memory (ICML 2024 · similarity 0.84)
- Fine-tuning Reinforcement Learning Models is Secretly a Forgetting Mitigation Problem (ICML 2024 · similarity 0.84)
- A Transfer Approach Using Graph Neural Networks in Deep Reinforcement Learning (AAAI 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
